How to stop runaway AI

How can humans retain power over more intelligent AI beings?
Dr. Stuart J. Russell is a Professor of Computer Science at UC Berkeley and has been studying the development of artificial intelligence for decades.

While he doesn’t think this latest crop of generative AI tools necessarily presents a significant threat to humanity, he does think it has helped to open the public’s eyes to the potential risks of more intelligent AI that could be coming in the future.

“They’re giving people now, in a very real sense, what would it be like if we had artificial general intelligence on tap available 24/7 to solve any problem that we might have. And they’re also seeing in a very visceral way that could present real risks,” Russell explained in a recent interview with Freethink.

As Russell argues in his book Human Compatible: Artificial Intelligence and the Problem of Control, we need to be appropriately concerned about the future threat of human-level, artificial general intelligence (AGI) which could pose an existential threat to humanity unless we can ensure that these systems remain aligned with human values and goals.

He contends that the standard approach to designing AI systems — in which machines are programmed to maximize some objective function — is fundamentally flawed because these machines don’t actually understand the world around them in any comprehensive way, a flaw that, in his mind, could lead to unintended and catastrophic failures if it can’t sufficiently anticipate the consequences of its own actions. Instead, he proposes a new approach to AI design in which machines are explicitly programmed to defer to humans in matters of value and to operate within a framework of uncertain and incomplete knowledge. By ensuring that AI systems are “human-compatible” in this way, Russell argues that we can harness the enormous potential of AI while minimizing the risk of catastrophic outcomes.

Related
Farmers can fight invasive insects with AI and a robotic arm
As the invasive spotted lanternfly threatens to expand its range, Carnegie Mellon researchers are developing a robot to fight back.
Google unveils AI try-on feature for shopping
Google’s AI-powered virtual try-on feature lets shoppers see what an article of clothing would look like on a wide range of models.
No, AI probably won’t kill us all – and there’s more to this fear campaign than meets the eye
A dose of scepticism is warranted when considering the AI doomsayer narrative — there are commercial incentives to manufacture fear of AI.
AI is riding to the rescue on wildfires
AI-powered systems designed to detect, confirm, and detail wildfires at the earliest possible time may help firefighters tame infernos in the West.
New “AI doctor” predicts risk of death with 85% accuracy
A new AI was able to make accurate predictions about patients’ risk of death, readmittance, and more by analyzing medical notes.
Subscribe to Freethink for more great stories